Automated Violence Recognition And Alert System Using Deep Learning

Violence in public and semi-public spaces poses a serious threat to safety, while continuous manual monitoring of surveillance cameras remains inefficient and error-prone. To address this challenge, this paper presents a real-time automated violence detection and alert system based on deep learning. The proposed system employs a YOLOv8-based convolutional neural network to analyze live CCTV streams as well as recorded video footage and identify violent or abnormal activities without human intervention. To improve reliability and reduce false alarms caused by sudden movements or background noise, a temporal confidence smoothing mechanism is applied across consecutive frames before confirming an event. Once violence is detected, the system automatically triggers an email alert to authorized personnel and logs event details in a centralized dashboard for monitoring and review. The backend is implemented using FastAPI with lightweight data storage, enabling efficient alert handling and visualization. Experimental evaluation shows that the system achieves reliable detection accuracy while operating at near real-time speeds of approximately 10–12 frames per second on CPU-only hardware. The proposed framework offers a practical, low-cost, and responsive solution for enhancing safety in modern video surveillance environments.